CourseLarge Language Models: The Complete Course · Module 12: Multimodal Models · part 88 of 103
Part 88 · Module 12: Multimodal Models

Incident: Fake receipts that read perfectly

9 min read·13 Oct 2026
Module 12 · Incident file · About 25 minutes

What happened

AI-MADE RECEIPTS, 2024 TO 2025 AppZen: none AI receipts 2024 AppZen: ~14% of fraud docs Sep 2025 FT reports a surge Oct 2025 Ramp: $1m+ in 90 days Oct 2025 13 Nov: ICAEW guidance
Between 2024 and September 2025 AI-made receipts went from none to about 14% of the fraudulent documents AppZen saw (red); ICAEW published guidance in November (green).

In October 2025 the Financial Times reported a surge in fake expense receipts made with AI image generators "from OpenAI and Google, among others", as Computerworld summarised it (Viktor Eriksson, "The latest genAI scam: fake expense receipts", 27 October 2025). The tools, it said, can create "extremely realistic fake receipts in just a few seconds without any specialized knowledge."

The numbers came from the companies that check expense claims. AppZen, an expense software provider, reported that AI-generated receipts "accounted for roughly 14% of all fraudulent documents submitted last month", and that "zero were submitted last year" (Daniella Genovese, Fox Business, 28 October 2025). Ramp's agents "flagged more than $1 million in fraudulent invoices in just 90 days." ICAEW, the accountancy body for England and Wales, placed the AppZen figure in September 2025 and passed on the FT's observation that "the launch of OpenAI's GPT-4o model last year coincided with a sharp rise in falsified receipts" (ICAEW Insights, "Expenses fraud: how to spot an AI-generated receipt", 13 November 2025).

WHAT THE REPORTS QUANTIFY AppZen, 2024 0% of fraudulent documents were AI-made AppZen, Sep 2025 about 14% Ramp, 90 days over $1 million in fraudulent invoices flagged
The reported figures (red) come from the vendors that caught the fakes; they describe flagged documents, not all fraud.

The cost was money paid on documents that described events that never happened. Arun Chauhan of Tenet Law told ICAEW: "These fakes are now so realistic that they can be very difficult to detect." Computerworld reported that detection tools "analyze metadata, patterns in file names and other details", but that "even metadata can be deleted."

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.